Add support for structured outputs #2
Description
OpenAI, Anthropic, Google, xAI all support structured output.
If the user provides a schema in the "Response Format" field of the Agent block, we should ensure that we are validating that schema and responding in the format that is expected by the user.
This is important because if the user declares this schema in a sub-block and references it in an API block that takes the output from an Agent block with the schema in its "Response Format" param, if the LLM responds with any other format the subsequent API call will also likely fail.